Linux HA Cluster in Vsphere across 2 VM Hosts

[JasonCal]

Hi All,

I have been in the process of building a Linux Cluster in a Vsphere utilizing SuSE's HAE product. This utilizes Pacemaker/Corosync for clustering resources. The main purpose is to build a File Server that use NFS to export files. The file server is on the smaller scale but would like the ability to scale out in terms of storage & performance if requirements change.

So far I've looked into DRBD and OCFS2 and built some test instances within these clusters, but found some limitations to our requirements mainly around snapshotting VMs. I've also looked into using VMware's Fault Tolerance in HA Mode and again found some limitations as well. So I am hoping someone could suggest something that perhaps I've missed? Are there any other options I am not considering?

So here's the requirements we are trying to satisfy

Requirement: 

• A Linux based NFS server
• Seamless redundancy between 2 VM’s
• Ability to provide High Availability over 2 ESXi Hosts.
• Ability to create snapshots for VMs
• Ability to do VDP backups
• Ability to support 2 CPUs (1 CPU may not be scalable)
• Storage capacity of approx. 150-200 GB for Data & 20 GB for OS

Option 1: 2 x SuSE in Linux HA Cluster (w/ OCFS2) VM’s w/ RDM mapping

• This was the option we were currently working with where we create an RDM to our SAN and share that mapping across multiple VM’s over 2 ESXi hosts
• Limitations:
  • Cannot backup any data nor do snapshots because of RDM
  • Cannot do snapshots of the OS disk because the VMs live on 2 different ESXi hosts and will require the Bus Sharing to be set to Physical to share the RDM across multiple VMs
  • Only way to make this work is to sacrifice HA by putting all VM's on a single host.

 

Option 2: 2 x SuSE in Linux HA Cluster (w/ DRBD) using Data stores

• All drives use their own Data Stores, but there is a shared disk to support STONITH. The Bus Sharing for that one controller is in physical
  mode.
• Limitations:
  • Old technology
  • Inability to scale well
  • Novell support requires a STONITH device to be enabled such as a shared disk to receive support for their HA product.  Therefore we are stuck with a shared disk in physical mode to have Availability across multiple ESXi hosts.

 

Option 3: ESXi HA Cluster using Fault Tolerance

• A single VM with Linux and enable Fault Tolerance option. Fault Tolerance would create an identical VM to provide redundancy and would continue to run seamlessly if the first VM ever lost its
  connection.

• Limitations
  • Requires 3 ESXi Hosts in HA Cluster mode
  • Cannot take snapshots
  • Fault Tolerance currently does not support more than 1 CPU

Thank you,

Jason

[JasonCal]

In addition to my question, I was doing a little research and was hoping to get some information about this VMware KB article I found regarding RDM compatibility modes.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2009226

It specifically mentions that Virtual Compatibility Mode for an RDM:

• If you are using a raw disk in virtual mode, you can realize the benefits of VMFS, such as advanced file locking for data protection and snapshots for streamlining development processes.
• You can use this mode for both  Cluster-in-a-box and cluster-across-boxes.

If I understand this correctly, it is saying I can use snapshots on VM if the RDM is in virtual compatibility mode and across 2 VM hosts. But having tried this, I found I still had to set Bus Sharing to Physical for both vm's to access the vmdk. This of course disables snapshots. Could someone help me understand this?

Thank you,

Jason

[ExpletiveDelete]

If you're using RDMs in a 'physical' mode (under SCSI bus sharing), then no, you can't snapshot the VM nor can it be vMotion'd while it its turned on. This is generally used when you've got a couple of SQL boxes that you're using MSCS for - that would generally be on separate hosts for HA/instance HA. The RDM itself is just a pointer file to the actual data LUN which is stationary.

Is this a production box? is there a reason (other than backups) that you need to snap-shot the box? If you're looking to backup the data, your best bet is to go with a client-based option that will take the data straight from the OS itself.

Also - consider vReplication. You can replicate the data based on increments as low as 15 minutes... no snapshots, great RPO/RTO options, and you can do it with a single vCenter, on the same or different site.

[JasonCal]

Thank you for your reply. This new proposed solution is currently in Dev.

There are two main purposes for backups/snapshots. One being around patching the OS .We want to ensure we can revert back incase an OS update causes an issue. The second being to secure our data on the server. This is how we have working in our current solution utilizing DRBD and using snapshots and VDP.

[ExpletiveDelete]

If those are hard requirements, I'd go with vSphere Replication. You can set up a replication to occur at specific times or manually.

VMware vSphere Replication Documentation

[JLackman]

With HA in 6.0, you can now do 4 cores...

[MRoushdy]

Hello,

I've done a Linux cluster on two VMs by sharing a virtual RDM disks, in this case you can snapshot. try it.